刑侦现场勘验图像具有图像质量差异大且多种类图像混合的特点，使得单一类型滤波器在低级视觉特征提取时存在特征定位准确度低、尺度难以估计、细节信息容易丢失的问题。不同类型滤波器在特征描述上具有各自优势并且信息互补，为此，本课题研究多类型滤波器协同的刑侦现勘图像低级视觉特征提取方法。具体从以下方面进行研究：(1)研究基于滤波器级协同和基于特征空间级协同技术，实现高效鲁棒的边缘和兴趣点特征提取方法; (2)针对彩色现勘图像，研究彩色视觉保持的向量排序方法和彩色变化最大化准则下最优分量融合方法，解决面向彩色现勘图像的多类型滤波器协同技术研究中的技术难点; (3)针对现勘图像中特有的多源图像和纹理类图像，分别利用统计学习理论和Bayesian估计方法实现对专有特征的精确建模和高效提取。课题的研究对未来构建全国高性能刑侦现勘系统提供了重要的理论和技术支撑。

The crime scene investigation (CSI) images have large difference in image quality and are rich in variety. In such environment, it would cause the problems of low accuracy in feature localization, hard choice for scale and high missing rate for image detail when the single filter is used for the feature extraction of CSI image. In fact, different types of filters have their own advantages but complementary information in feature description. Thus, this project studies the feature extraction method of crime scene investigation image by collaborating with multiple types of filters. The detailed researches include the following three aspects, (1) The filter level collaboration and feature space level collaboration techniques are studied for the effective and robust edge and interest point extraction in CSI. (2) For the color CSI images, we focus on the research of the color visual distance preserving vector ordering technique and optimal multiple components fusion technique under color change maximization. The solution of these two techniques helps the feature extraction method by collaborating with multiple types of filters for color CSI image. (3) For the exclusive multi-sensor images and texture-like images in CSI images, the statistical learning theory and Bayesian estimation method are used for the accurate modeling and effective extraction of exclusive feature in images. In conclusion, this project will provide important theory and technique support for establishing the national CSI system with high performance in the future.